One of the things shaping the big picture here is fairness. Is it fair for insurers to collect and use such data? Of course, say insurers. Of course not, say consumers. How dare they spy on what I’m doing at home, complains the latter.
Remember that you need to think about fairness on two levels in situations like this. Firstly, is the underlying rationale fair? And secondly, is it being put into practice fairly. In situations like this, fairness can often be working on both levels at the same time.
An interesting example of this move to use internal risk data as well as external risk data, and the fairness debate that emerges, comes from the US state of California, where market conditions are not particularly stable at the moment. Insurers there have lowered their risk appetite and invested in technology to strip out policies found to be below their new levels of tolerance.
It appears that if you have a back garden that looks cluttered, or a roof that looks a bit old, or a swimming pool that’s been drained, then these are reasons for some Californian insurers not to offer renewal, even if you’ve had a long and claims free history with them.
Automated Image Analysis
Insurers appear to be using a largely automated process that relies on image data obtained on an area by area basis through third parties. Drones, fixed wing planes and even satellites are being used to collect this image data.
Insurers are refusing to go into any detail about their decision to non-renew, and changing a decision seems to be pretty rare. When one policyholder was told that their roof had “exceeded its useful life”, it took some recent invoices for the insurer to change their mind, I suspect largely for PR reasons.
Two market trends are at play here. The first is about insurers stripping their portfolios in certain parts of the US of anything other than the very best policies. You may think this is down an increase in climate related catastrophes, but that’s not always the case. It’s more about insurers wanting to put in significant price increases and finding that regulators are not going along with this.
The second market trend is the move from personalised underwriting to hyper-personalised underwriting. It’s no longer just about where your home is, but also about what your home represents as well. On the surface, this would seem to be about the property, but in reality, what is of interest to the insurer is the character of the policyholder.
A cluttered back garden indicates a policyholder who doesn’t take enough care, even though the person just liked renovating his old car. An emptied swimming pool was also seen in terms of lack of care, (the insurer talked about “deferred maintenance”), even though the owners no longer used it and found refilling it expensive.
Efficiency, not Accuracy
The second thing shaping the big picture here is technology. On one level, artificial intelligence is being used to analyse image data and deliver quantitative judgements as to the condition of the insured property. On the second level is some form of AI model being used to deliver a profitable portfolio based upon a detailed risk appetite and applied to a large portfolio of policies. Out of this comes a list of policies not to be renewed, and no doubt performance management is being used to back this up.
The most that an insurer will say about non-renewal is couched in broad terms that the public think might have some bearing on risk. Deferred maintenance can be a risk factor, but difficult to logically link to an empty swimming pool. Is this a case then of ‘I do not want to renew this policy so find me a reason not to’?
What we see here here is not artificial intelligence as an objective and accurate way of underwriting policies, but artificial intelligence as an efficient but still rather arbitrary way of putting a detailed risk appetite into operation.
The wider problem here is that more and more insurers start to chase a smaller and smaller pool of low risk policies. And in doing so, they will seek to extract more and more data about each policy in order to stay just far enough ahead of the others to make a profit. Until of course, a competitor does pretty much the same but incrementally better.
And as more and more insurers equip themselves with more and more similar AI risk models, then the sources of competitive advantage from AI become rarer and rarer. One or two of the more tuned in insurers have realised this and started to think differently, but at the moment, most in the market are following the herd.
Not Just California
It’s only California, not the whole of the US, some of you may be thinking. It’s worth remembering however that California on its own is the fourth biggest insurance market in the world. This then is the state’s equivalent of the flood underwriting situation that emerged in the UK in the mid 2010s. In California’s case however, a range of perils is involved and a lot more data and analytics.
At some point then, all those policyholders, consumer groups, state insurance commissioners and attorney generals will start to push back, unwilling to accept the impact that personalisation and hyper-personalisation are having on insurance provision. I’m pretty sure of how this push back will be shaped, for one or two signs of it are emerging.
What we have then, to use the language from a panel I was on at a 2017 EU policy conference, is a market that thinks it’s moving onto a digital superhighway, but which everyone else is experiencing as heading towards a cliff. What happened in the UK, and is likely to happen in the US, was a great focussing of political minds on a problem that hadn’t existed a decade ago, but was now a problem affecting voters across the political spectrum.
It’s not for nothing that insurance is described as a political technology.